Priority Inheritance with Backtracking for Iterative Multi-Agent Path Finding

Total Page:16

File Type:pdf, Size:1020Kb

Priority Inheritance with Backtracking for Iterative Multi-Agent Path Finding Priority Inheritance with Backtracking for Iterative Multi-agent Path Finding Keisuke Okumura1 , Manao Machida2 , Xavier Defago´ 1 and Yasumasa Tamura1 1Tokyo Institute of Technology 2NEC Corporation [email protected], [email protected], fdefago, [email protected] Abstract That is, whenever an agent reaches a goal, it receives a new one. This rules out any simple adaptation of offline The Multi-agent Path Finding (MAPF) problem and computationally-intensive optimal solutions due to pro- consists in all agents having to move to their own hibitive computations, and new goals typically appearing at destinations while avoiding collisions. In practical runtime. Furthermore, centralized solutions being inherently applications to the problem, such as for navigation problematic for systems with many agents due to scalabil- in an automated warehouse, MAPF must be solved ity concerns, decoupled algorithms such as prioritized route iteratively. We present here a novel approach to it- planning are adequate, especially if fully decentralized. De- erative MAPF, that we call Priority Inheritance with centralized solutions are highly attractive to multi-agent/robot Backtracking (PIBT). PIBT gives a unique priority systems for many reasons, including a higher potential for to each agent every timestep, so that all movements robustness and fault-tolerance, better scalability, and lower are prioritized. Priority inheritance, which aims at production cost [Yan et al., 2013]. Finally, implementations dealing effectively with priority inversion in path without global communication are preferable since they pro- adjustment within a small time window, can be ap- vide better potential for scalability and concurrency. plied iteratively and a backtracking protocol pre- iterative vents agents from being stuck. We prove that, re- In this paper, we present a novel algorithm for Multi-agent Path Finding Priority gardless of their number, all agents are guaranteed (iterative MAPF), called Inheritance with Backtracking to reach their destination within finite time, when (PIBT), which focuses on the the environment is a graph such that all pairs of ad- adjacent movements of multiple agents based on prioritized jacent nodes belong to a simple cycle of length 3 planning in a short time window. Priority inheritance is a or more (e.g., biconnected). Our implementation well-known approach to deal effectively with priority inver- [ et al. ] of PIBT can be fully decentralized without global sion in real-time systems Sha , 1990 , and is applied communication. Experimental results over vari- here to path adjustment. When a low-priority agent X im- ous scenarios confirm that PIBT is adequate both pedes the movement of a higher-priority agent Y, agent X for finding paths in large environments with many temporarily inherits the higher-priority of Y. To avoid a sit- agents, as well as for conveying packages in an au- uation where agents are stuck waiting, priority inheritance is tomated warehouse. executed in combination with a backtracking protocol. Since PIBT assumes that agents can only communicate when lo- cated within two hops of each other (e.g., Manhattan distance 1 Introduction 2 in a grid environment), it can be implemented in a fully In systems using physical moving agents, it is essential to al- decentralized way and inherits the above characteristics. arXiv:1901.11282v3 [cs.MA] 23 Jun 2019 low agents to reach their own destinations smoothly without Our main contributions are two-folds: 1) we propose an al- collisions, by providing valid paths while minimizing excess gorithm ensuring that every agent always reaches its destina- travel time. This problem, known as Multi-agent Path Find- tion within finite time as long as the environment satisfies the ing (MAPF), is however computationally difficult due to the condition that all pairs of adjacent nodes belong to a simple search space growing exponentially as the number of agents cycle of length 3 or more (includes undirected biconnected increases. MAPF is an important problem due to its appli- graphs); and 2) we evaluate that algorithm in various envi- cations in various domains, e.g., traffic control [Dresner and ronments, showing its practicality. In particular, experimental Stone, 2008], automated warehouse [Wurman et al., 2008], results over various scenarios confirm its adequateness both or airport surface operation [Morris et al., 2016], etc. for finding paths in large environments with many agents, as Previous research on MAPF focuses on solving a “one- well as for conveying packages in an automated warehouse. shot” version of the problem, i.e., to make agents reach their The paper is organized as follows. Section 2 reviews ex- goal from their initial position only once. In practical ap- isting algorithms for MAPF and a variant called Multi-agent plications, such as conveying packages in a warehouse [Ma Pickup and Delivery (MAPD). Section 3 defines iterative et al., 2017], MAPF must however be solved iteratively. MAPF, which provides an abstract framework for multiple moving agents. Section 4 presents the PIBT algorithm and its a delivery location. The paper proposes two decoupled algo- theoretical analysis. Section 5 presents empirical results on rithms based on HCA∗ for MAPD, called respectively Token both path finding (MAPF) and pickup and delivery (MAPD). Passing (TP) and Token Passing with Task Swap. These algo- Section 6 concludes the paper and discusses future work. rithms can easily be adapted to be decentralized but require a certain amount of non-task endpoints where agents do not 2 Related Work block other agents’ motion. Many complete MAPF algorithms exist, such as A∗ with Op- erator Decomposition [Standley, 2010], Enhanced Partial Ex- pansion A∗ [Goldenberg et al., 2014], Increasing Cost Tree 3 Problem Definition Search [Sharon et al., 2013], Conflict-based Search [Sharon ∗ et al., 2015],M [Wagner and Choset, 2015], etc. The problem of iterative Multi-agent Path Finding (itera- Finding an optimal solution is however NP-hard [Yu and tive MAPF) is a generalization of problems addressing mul- LaValle, 2013], and optimal algorithms do not scale in the tiple moving agents, including both Multi-agent Path Find- number of agents and are too costly for iterative use, hence ing (MAPF) and Multi-agent Pickup and Delivery (MAPD). the need for sub-optimal solvers, e.g., FAR [Wang et al., Since iterative MAPF is an abstract model we do not intend to 2008], MAPP [Wang and Botea, 2011], Tree-based Agent solve it directly, rather, it is necessary to embody task creation Swapping Strategy [Khorshid et al., 2011], BIBOX [Surynek, according to target domains. 2009], CBS variants [Barer et al., 2014; Cohen et al., 2016]. Push and Swap/Rotate [Luna and Bekris, 2011; de Wilde et al., 2013], which partly inspired our proposal, are sub- The system consists of a set of agents, A = fa1; : : : ; ang, optimal centralized approaches that allow one agent to push and an environment given as a graph G = (V; E), where another away from its path. However, they only allow a sin- agents occupy nodes in V and move along edges in E. Con- gle agent or a pair of agents to move at each timestep. En- sidering practical situations, G must be a simple (neither hanced Push and Swap include Parallel Push and Swap [Sajid loops nor multi-edges) and strongly-connected (every node et al., 2012] where all agents can move simultaneously, or is reachable from every other node) directed graph. This in- Push-Swap-Wait [Wiktor et al., 2014] which takes a decen- cludes all simple undirected graphs that are connected. tralized approach in narrow passages. DisCoF [Zhang et al., Let vi(t) denote the node occupied by agent ai at discrete 2016] combines decoupled prioritized planning with fallback time t. The initial position of agent ai is vi(0) and given to coupled push and swap in case of conflicts. PIBT, the pro- as input. At each timestep, ai selects one node vi(t + 1) 2 posed method, can be seen as a combination of safe “push” fvjv 2 V; (vi(t); v) 2 Eg [ fvi(t)g as its location for the operations, thanks to backtracking and dynamic priorities. next timestep. Agents must avoid 1) collision: vi(t) 6= vj(t); Prioritized planning is computationally cheap and hence and 2) intersection with others: vi(t) 6= vj(t + 1) _ vj(t + ∗ ∗ attractive for MAPF. Hierarchical Cooperative A (HCA ) 1) 6= vj(t). We do not prohibit rotations, so vi(t + 1) = ∗ ∗ and Windowed HCA (WHCA ) [Silver, 2005] are decou- vj(t)^vj(t+1) = vk(t)^· · ·^vl(t+1) = vi(t) is possible. pled approaches in that they plan a path for each agent Let Γ = fτ1; τ2;::: g be a dynamic set of tasks with new one after the other while avoiding collisions with previously tasks being added over time, i.e., not all tasks are known computed paths. WHCA∗ uses a limited lookahead win- ∗ initially. A task is defined as a finite set of goals τj = dow. Our proposal, PIBT, is based on WHCA with a win- fg ; g ; : : : ; g g where g 2 V , possibly with dependencies dow size of one. Decentralized solutions, i.e., where each 1 2 m k as a partial order on gk. Let tj be the timestep when τj is agent computes its own path based on information from added to Γ. An agent is free when it has no assigned task. A other agents, are inherently decoupled. Decentralized solu- task τj 2 Γ can only be assigned to free agents. When τj is tions for MAPF [Velagapudi et al., 2010; Cˇ ap´ et al., 2015; assigned to ai, ai starts visiting goals in τj. When all goals Chouhan and Niyogi, 2015] are hence usually prioritized. have been visited, τj is completed and ai becomes free. The negotiation process for ordering priorities studied by Azarm et al.
Recommended publications
  • Lecture 4 Dynamic Programming
    1/17 Lecture 4 Dynamic Programming Last update: Jan 19, 2021 References: Algorithms, Jeff Erickson, Chapter 3. Algorithms, Gopal Pandurangan, Chapter 6. Dynamic Programming 2/17 Backtracking is incredible powerful in solving all kinds of hard prob- lems, but it can often be very slow; usually exponential. Example: Fibonacci numbers is defined as recurrence: 0 if n = 0 Fn =8 1 if n = 1 > Fn 1 + Fn 2 otherwise < ¡ ¡ > A direct translation in:to recursive program to compute Fibonacci number is RecFib(n): if n=0 return 0 if n=1 return 1 return RecFib(n-1) + RecFib(n-2) Fibonacci Number 3/17 The recursive program has horrible time complexity. How bad? Let's try to compute. Denote T(n) as the time complexity of computing RecFib(n). Based on the recursion, we have the recurrence: T(n) = T(n 1) + T(n 2) + 1; T(0) = T(1) = 1 ¡ ¡ Solving this recurrence, we get p5 + 1 T(n) = O(n); = 1.618 2 So the RecFib(n) program runs at exponential time complexity. RecFib Recursion Tree 4/17 Intuitively, why RecFib() runs exponentially slow. Problem: redun- dant computation! How about memorize the intermediate computa- tion result to avoid recomputation? Fib: Memoization 5/17 To optimize the performance of RecFib, we can memorize the inter- mediate Fn into some kind of cache, and look it up when we need it again. MemFib(n): if n = 0 n = 1 retujrjn n if F[n] is undefined F[n] MemFib(n-1)+MemFib(n-2) retur n F[n] How much does it improve upon RecFib()? Assuming accessing F[n] takes constant time, then at most n additions will be performed (we never recompute).
    [Show full text]
  • Exhaustive Recursion and Backtracking
    CS106B Handout #19 J Zelenski Feb 1, 2008 Exhaustive recursion and backtracking In some recursive functions, such as binary search or reversing a file, each recursive call makes just one recursive call. The "tree" of calls forms a linear line from the initial call down to the base case. In such cases, the performance of the overall algorithm is dependent on how deep the function stack gets, which is determined by how quickly we progress to the base case. For reverse file, the stack depth is equal to the size of the input file, since we move one closer to the empty file base case at each level. For binary search, it more quickly bottoms out by dividing the remaining input in half at each level of the recursion. Both of these can be done relatively efficiently. Now consider a recursive function such as subsets or permutation that makes not just one recursive call, but several. The tree of function calls has multiple branches at each level, which in turn have further branches, and so on down to the base case. Because of the multiplicative factors being carried down the tree, the number of calls can grow dramatically as the recursion goes deeper. Thus, these exhaustive recursion algorithms have the potential to be very expensive. Often the different recursive calls made at each level represent a decision point, where we have choices such as what letter to choose next or what turn to make when reading a map. Might there be situations where we can save some time by focusing on the most promising options, without committing to exploring them all? In some contexts, we have no choice but to exhaustively examine all possibilities, such as when trying to find some globally optimal result, But what if we are interested in finding any solution, whichever one that works out first? At each decision point, we can choose one of the available options, and sally forth, hoping it works out.
    [Show full text]
  • Backtrack Parsing Context-Free Grammar Context-Free Grammar
    Context-free Grammar Problems with Regular Context-free Grammar Language and Is English a regular language? Bad question! We do not even know what English is! Two eggs and bacon make(s) a big breakfast Backtrack Parsing Can you slide me the salt? He didn't ought to do that But—No! Martin Kay I put the wine you brought in the fridge I put the wine you brought for Sandy in the fridge Should we bring the wine you put in the fridge out Stanford University now? and University of the Saarland You said you thought nobody had the right to claim that they were above the law Martin Kay Context-free Grammar 1 Martin Kay Context-free Grammar 2 Problems with Regular Problems with Regular Language Language You said you thought nobody had the right to claim [You said you thought [nobody had the right [to claim that they were above the law that [they were above the law]]]] Martin Kay Context-free Grammar 3 Martin Kay Context-free Grammar 4 Problems with Regular Context-free Grammar Language Nonterminal symbols ~ grammatical categories Is English mophology a regular language? Bad question! We do not even know what English Terminal Symbols ~ words morphology is! They sell collectables of all sorts Productions ~ (unordered) (rewriting) rules This concerns unredecontaminatability Distinguished Symbol This really is an untiable knot. But—Probably! (Not sure about Swahili, though) Not all that important • Terminals and nonterminals are disjoint • Distinguished symbol Martin Kay Context-free Grammar 5 Martin Kay Context-free Grammar 6 Context-free Grammar Context-free
    [Show full text]
  • Backtracking / Branch-And-Bound
    Backtracking / Branch-and-Bound Optimisation problems are problems that have several valid solutions; the challenge is to find an optimal solution. How optimal is defined, depends on the particular problem. Examples of optimisation problems are: Traveling Salesman Problem (TSP). We are given a set of n cities, with the distances between all cities. A traveling salesman, who is currently staying in one of the cities, wants to visit all other cities and then return to his starting point, and he is wondering how to do this. Any tour of all cities would be a valid solution to his problem, but our traveling salesman does not want to waste time: he wants to find a tour that visits all cities and has the smallest possible length of all such tours. So in this case, optimal means: having the smallest possible length. 1-Dimensional Clustering. We are given a sorted list x1; : : : ; xn of n numbers, and an integer k between 1 and n. The problem is to divide the numbers into k subsets of consecutive numbers (clusters) in the best possible way. A valid solution is now a division into k clusters, and an optimal solution is one that has the nicest clusters. We will define this problem more precisely later. Set Partition. We are given a set V of n objects, each having a certain cost, and we want to divide these objects among two people in the fairest possible way. In other words, we are looking for a subdivision of V into two subsets V1 and V2 such that X X cost(v) − cost(v) v2V1 v2V2 is as small as possible.
    [Show full text]
  • Module 5: Backtracking
    Module-5 : Backtracking Contents 1. Backtracking: 3. 0/1Knapsack problem 1.1. General method 3.1. LC Branch and Bound solution 1.2. N-Queens problem 3.2. FIFO Branch and Bound solution 1.3. Sum of subsets problem 4. NP-Complete and NP-Hard problems 1.4. Graph coloring 4.1. Basic concepts 1.5. Hamiltonian cycles 4.2. Non-deterministic algorithms 2. Branch and Bound: 4.3. P, NP, NP-Complete, and NP-Hard 2.1. Assignment Problem, classes 2.2. Travelling Sales Person problem Module 5: Backtracking 1. Backtracking Some problems can be solved, by exhaustive search. The exhaustive-search technique suggests generating all candidate solutions and then identifying the one (or the ones) with a desired property. Backtracking is a more intelligent variation of this approach. The principal idea is to construct solutions one component at a time and evaluate such partially constructed candidates as follows. If a partially constructed solution can be developed further without violating the problem’s constraints, it is done by taking the first remaining legitimate option for the next component. If there is no legitimate option for the next component, no alternatives for any remaining component need to be considered. In this case, the algorithm backtracks to replace the last component of the partially constructed solution with its next option. It is convenient to implement this kind of processing by constructing a tree of choices being made, called the state-space tree. Its root represents an initial state before the search for a solution begins. The nodes of the first level in the tree represent the choices made for the first component of a solution; the nodes of the second level represent the choices for the second component, and soon.
    [Show full text]
  • CS/ECE 374: Algorithms & Models of Computation
    CS/ECE 374: Algorithms & Models of Computation, Fall 2018 Backtracking and Memoization Lecture 12 October 9, 2018 Chandra Chekuri (UIUC) CS/ECE 374 1 Fall 2018 1 / 36 Recursion Reduction: Reduce one problem to another Recursion A special case of reduction 1 reduce problem to a smaller instance of itself 2 self-reduction 1 Problem instance of size n is reduced to one or more instances of size n − 1 or less. 2 For termination, problem instances of small size are solved by some other method as base cases. Chandra Chekuri (UIUC) CS/ECE 374 2 Fall 2018 2 / 36 Recursion in Algorithm Design 1 Tail Recursion: problem reduced to a single recursive call after some work. Easy to convert algorithm into iterative or greedy algorithms. Examples: Interval scheduling, MST algorithms, etc. 2 Divide and Conquer: Problem reduced to multiple independent sub-problems that are solved separately. Conquer step puts together solution for bigger problem. Examples: Closest pair, deterministic median selection, quick sort. 3 Backtracking: Refinement of brute force search. Build solution incrementally by invoking recursion to try all possibilities for the decision in each step. 4 Dynamic Programming: problem reduced to multiple (typically) dependent or overlapping sub-problems. Use memoization to avoid recomputation of common solutions leading to iterative bottom-up algorithm. Chandra Chekuri (UIUC) CS/ECE 374 3 Fall 2018 3 / 36 Subproblems in Recursion Suppose foo() is a recursive program/algorithm for a problem. Given an instance I , foo(I ) generates potentially many \smaller" problems. If foo(I 0) is one of the calls during the execution of foo(I ) we say I 0 is a subproblem of I .
    [Show full text]
  • Toward a Model for Backtracking and Dynamic Programming
    TOWARD A MODEL FOR BACKTRACKING AND DYNAMIC PROGRAMMING Michael Alekhnovich, Allan Borodin, Joshua Buresh-Oppenheim, Russell Impagliazzo, Avner Magen, and Toniann Pitassi Abstract. We propose a model called priority branching trees (pBT ) for backtrack- ing and dynamic programming algorithms. Our model generalizes both the priority model of Borodin, Nielson and Rackoff, as well as a simple dynamic programming model due to Woeginger, and hence spans a wide spectrum of algorithms. After witnessing the strength of the model, we then show its limitations by providing lower bounds for algorithms in this model for several classical problems such as Interval Scheduling, Knapsack and Satisfiability. Keywords. Greedy Algorithms, Dynamic Programming, Models of Computation, Lower Bounds. Subject classification. 68Q10 1. Introduction The “Design and Analysis of Algorithms” is a basic component of the Computer Science Curriculum. Courses and texts for this topic are often organized around a toolkit of al- gorithmic paradigms or meta-algorithms such as greedy algorithms, divide and conquer, dynamic programming, local search, etc. Surprisingly (as this is often the main “theory course”), these algorithmic paradigms are rarely, if ever, precisely defined. Instead, we provide informal definitional statements followed by (hopefully) well chosen illustrative examples. Our informality in algorithm design should be compared to computability the- ory where we have a well accepted formalization for the concept of an algorithm, namely that provided by Turing machines and its many equivalent computational models (i.e. consider the almost universal acceptance of the Church-Turing thesis). While quantum computation may challenge the concept of “efficient algorithm”, the benefit of having a well defined concept of an algorithm and a computational step is well appreciated.
    [Show full text]
  • Tree Search Backtracking Search Backtracking Search
    Search: tree search • Now let's put modeling aside and suppose we are handed a search problem. How do we construct an algorithm for finding a minimum cost path (not necessarily unique)? Backtracking search • We will start with backtracking search, the simplest algorithm which just tries all paths. The algorithm is called recursively on the current state s and the path leading up to that state. If we have reached a goal, then we can update the minimum cost path with the current path. Otherwise, we consider all possible actions a from state s, and recursively search each of the possibilities. • Graphically, backtracking search performs a depth-first traversal of the search tree. What is the time and memory complexity of this algorithm? • To get a simple characterization, assume that the search tree has maximum depth D (each path consists of D actions/edges) and that there are b available actions per state (the branching factor is b). • It is easy to see that backtracking search only requires O(D) memory (to maintain the stack for the recurrence), which is as good as it gets. • However, the running time is proportional to the number of nodes in the tree, since the algorithm needs to check each of them. The number 2 D bD+1−1 D of nodes is 1 + b + b + ··· + b = b−1 = O(b ). Note that the total number of nodes in the search tree is on the same order as the [whiteboard: search tree] number of leaves, so the cost is always dominated by the last level.
    [Show full text]
  • CSCI 742 - Compiler Construction
    CSCI 742 - Compiler Construction Lecture 12 Recursive-Descent Parsers Instructor: Hossein Hojjat February 12, 2018 Recap: Predictive Parsers • Predictive Parser: Top-down parser that looks at the next few tokens and predicts which production to use • Efficient: no need for backtracking, linear parsing time • Predictive parsers accept LL(k) grammars • L means “left-to-right” scan of input • L means leftmost derivation • k means predict based on k tokens of lookahead 1 Implementations Analogous to lexing: Recursive descent parser (manual) • Each non-terminal parsed by a procedure • Call other procedures to parse sub-nonterminals recursively • Typically implemented manually Table-driven parser (automatic) • Push-down automata: essentially a table driven FSA, plus stack to do recursive calls • Typically generated by a tool from a grammar specification 2 Making Grammar LL • Recall the left-recursive grammar: S ! S + E j E E ! num j (E) • Grammar is not LL(k) for any number of k • Left-recursion elimination: rewrite left-recursive productions to right-recursive ones S ! EE0 S ! S + E j E E0 ! +EE0 j E ! num j (E) E ! num j (E) 3 Making Grammar LL • Recall the grammar: S ! E + E j E E ! num j (E) • Grammar is not LL(k) for any number of k • Left-factoring: Factor common prefix E, add new non-terminal E0 for what follows that prefix S ! EE0 S ! E + E j E E0 ! +E j E ! num j (E) E ! num j (E) 4 Parsing with new grammar S ! EE0 E0 ! +E j E ! num j (E) Partly-derived String Lookahead parsed part unparsed part S ( (num)+num ) E E0 ( (num)+num ) (E)
    [Show full text]
  • CSE 100: GRAPH SEARCH Announcements
    CSE 100: GRAPH SEARCH Announcements • PA3 released • Checkpoint due Tuesday, May 5 @10:00pm • Final submission due Thursday, May 14 @10:00PM Start early! Start early! Start early! Start early! Start early! I don’t mean to nag but…. Start early, Start early… The labs, they get jammed if you don’t …. Start early! The tutors, they can’t swim when the tide of students crash in, so… Start early! Start early! Post lecture is perfect! You will love PA3, It has a mi-ni-mum spanning tree, oh sooo pretty, you’ll see.. If you staaaaaaart earrrrrrlyyyyyyyy…… today 3 Generic approach to graph search Generic Goals: • Find everything that can be explored • Don’t explore anything twice V0 V1 V3 V4 V2 We will look at different Graph search algorithms • Depth First Search (very useful for PA4) • Breadth First Search • Best-First Search (for weighted graphs) (very useful for PA3) 4 Depth First Search for Graph Traversal • Search as far down a single path as possible before backtracking V0 V1 V3 V4 V5 V2 5 Depth First Search for Graph Traversal • Search as far down a single path as possible before backtracking V0 V1 V3 V4 V5 V2 Assuming DFS chooses the lower number node to explore first, in what order does DFS visit the nodes in this graph? A. V0, V1, V2, V3, V4, V5 B. V0, V1, V3, V4, V2, V5 C. V0, V1, V3, V2, V4, V5 D. V0, V1, V2, V4, V5, V3 Shortest Path Algorithms Finding the shortest route from one city to another is a natural applicaon of graph algorithms! (Of course there are many other examples) Unweighted Shortest Path • Input: an unweighted directed graph G = (V, E) and a source vertex s in V • Output: for each vertex v in V, a representation of the shortest path in G that starts in s and ends at v and consists of the minimum number of edges compared to any other path from s to v Depth First Search for Graph Traversal • Search as far down a single path as possible before backtracking V0 V1 V3 V4 V5 V2 Does DFS always find the shortest path between nodes? A.
    [Show full text]
  • Lecture 9: Games I
    Lecture 9: Games I CS221 / Spring 2018 / Sadigh Course plan Search problems Markov decision processes Constraint satisfaction problems Adversarial games Bayesian networks Reflex States Variables Logic "Low-level intelligence" "High-level intelligence" Machine learning CS221 / Spring 2018 / Sadigh 1 • This lecture will be about games, which have been one of the main testbeds for developing AI programs since the early days of AI. Games are distinguished from the other tasks that we've considered so far in this class in that they make explicit the presence of other agents, whose utility is not generally aligned with ours. Thus, the optimal strategy (policy) for us will depend on the strategies of these agents. Moreover, their strategies are often unknown and adversarial. How do we reason about this? A simple game Example: game 1 You choose one of the three bins. I choose a number from that bin. Your goal is to maximize the chosen number. A B C -50 50 1 3 -5 15 CS221 / Spring 2018 / Sadigh 3 • Which bin should you pick? Depends on your mental model of the other player (me). • If you think I'm working with you (unlikely), then you should pick A in hopes of getting 50. If you think I'm against you (likely), then you should pick B as to guard against the worst case (get 1). If you think I'm just acting uniformly at random, then you should pick C so that on average things are reasonable (get 5 in expectation). Roadmap Games, expectimax Minimax, expectiminimax Evaluation functions Alpha-beta pruning CS221 / Spring 2018 / Sadigh 5 Game tree Key idea: game tree Each node is a decision point for a player.
    [Show full text]
  • Lecture 3: Backtracking
    Lecture 3: Backtracking Last updated Feb 1, 2021 References: Algorithms, Jeff Erickson, Chapter 2 Backtracking 2/26 Backtracking is a particular recursive strategy which constructs a slution incrementally, one piece at a time. Think the solution as a vector, and backtracking computes one ele- ment at a time. Whenever the algorithm needs to decide between multiple alterna- tives, it systematically evaluates every alternative and choose the best. N-Queens 3/26 On a 8x8 chessboard, place 8 queens such that no two queens attack each other. (no two queens in the same row, column, or diagonal). Find one solution, or all the solutions, or count the number of solutions. Backtracking strategy: For top row to bottom row, try to place one queen per row at every pos- sible position. Key data structure: partial solution. N-Queen Backtracking 4/26 The backtracking algorithm given by Gauss is essentially: PlaceQueens(Q[1::r 1]; r): if r > n print Q[1¡::n] # solution complete else for j 1 to n #consider position j in row r le gal true for i 1 to r 1 #checking if j is legal if (Q[i] =¡j) or (Q[i]=j+r-i) or (Q[i]=j-r+i) legal=false if legal Q.push(j); #make move PlaceQueens(Q[1::r]; r + 1) #accept j, recurse! Q.pop(); #unmake move N-Queen Backtracking 5/26 It's equivalent to depth-first search of this tree: Recursion tree: Node: legal partial solution. Edge: recursive calls Leaves: partial solu- tions that cannot be extended (deadend, or complete solution) N-Queens Implementation 6/26 See the animation page: http://www2.cs.uh.edu/~panruowu/ 2021s_cosc3320/nqueen.html To call the recursive backtracking routine, simply do: Game 7/26 Consider a simple two-player game on nxn square grid: Rules: Red/Green player takes turns Can move only horizon- tally (red) or vertically (green) Can move one step, or jump over one.
    [Show full text]